diff options
author | Nicola Corna <nicola@corna.info> | 2016-11-16 08:57:15 +0100 |
---|---|---|
committer | Martin Roth <martinroth@google.com> | 2016-12-27 18:18:04 +0100 |
commit | 76f8dbc4f70469e0b5913796d6130e85df1cdd77 (patch) | |
tree | e28af65fa0a0c8134840d2de087b96dfa0794fae | |
parent | aa5c8f78f98614130cfb4802faa30b6a6b92e815 (diff) |
device/dram/ddr3: add FTB timings
SPD revision 1.1 introduced FTB timings, an extra set of SPD values that
specify a more precise tCKmin, tAAmin, tRCDmin, tRPmin and tRCmin.
For backwards compatibility, the MTB is usually rounded up and the FTB
part is negative. For this reason some memories were not set up optimally,
as the FTB part was ignored and the resulting timing wasn't set to the
minimum value.
The tests were performed on a Lenovo X220 with two Micron 8KTF51264HZ-1G9E
(1866 MHz): reading only the MTB part, coreboot reports a tCKmin of
1.125 ns, corresponding to a working frequency of 800 MHz; with the
additional tCKmin FTB part (-0.054 ns) the new (rounded) value is
1.070 ns, valid for a 933 MHz operation.
Tested also with Ballistix DDR3-1866 SODIMM on Lenovo T420: the memory is
now detected as DDR3-1866 instead of DDR3-1600.
Some manufacturers (like Micron) seems to expect a small rounding on the
timings, so a nearest-value rounding is performed. If this assumption
isn't correct, an error up to ~2 ps can be committed, which is low enough
to be safely ignored.
Change-Id: Ib98f2e70820f207429d04ca6421680109a81f457
Signed-off-by: Nicola Corna <nicola@corna.info>
Reviewed-on: https://review.coreboot.org/17476
Tested-by: build bot (Jenkins)
Reviewed-by: Patrick Rudolph <siro@das-labor.org>
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
-rw-r--r-- | src/device/dram/ddr3.c | 37 |
1 files changed, 30 insertions, 7 deletions
diff --git a/src/device/dram/ddr3.c b/src/device/dram/ddr3.c index 6e5c541a85..db661a9886 100644 --- a/src/device/dram/ddr3.c +++ b/src/device/dram/ddr3.c @@ -129,9 +129,10 @@ int spd_decode_ddr3(dimm_attr * dimm, spd_raw_data spd) { int ret; u16 crc, spd_crc; - u8 ftb_divisor, ftb_dividend, capacity_shift, bus_width; + u8 capacity_shift, bus_width; u8 reg8; u32 mtb; /* medium time base */ + u32 ftb; /* fine time base */ unsigned int val, param; ret = SPD_STATUS_OK; @@ -263,12 +264,6 @@ int spd_decode_ddr3(dimm_attr * dimm, spd_raw_data spd) dimm->size_mb = ((1 << (capacity_shift + (25 - 20))) * bus_width * dimm->ranks) / dimm->width; - /* Fine Timebase (FTB) Dividend/Divisor */ - /* Dividend */ - ftb_dividend = (spd[9] >> 4) & 0x0f; - /* Divisor */ - ftb_divisor = spd[9] & 0x0f; - /* Medium Timebase = * Medium Timebase (MTB) Dividend / * Medium Timebase (MTB) Divisor */ @@ -301,6 +296,34 @@ int spd_decode_ddr3(dimm_attr * dimm, spd_raw_data spd) /* Minimum Four Activate Window Delay Time (tFAWmin) */ dimm->tFAW = (((spd[28] & 0x0f) << 8) + spd[29]) * mtb; + printram(" FTB timings :"); + /* FTB is introduced in SPD revision 1.1 */ + if (spd[1] >= 0x11 && spd[9] & 0x0f) { + printram(" yes\n"); + + /* Fine timebase (1/256 ps) = + * Fine Timebase (FTB) Dividend / + * Fine Timebase (FTB) Divisor */ + ftb = (((u16) spd[9] & 0xf0) << 4) / (spd[9] & 0x0f); + + /* SPD recommends to round up the MTB part and use a negative + * FTB, so a negative rounding should be always safe */ + + /* SDRAM Minimum Cycle Time (tCKmin) correction */ + dimm->tCK += (s32)((s8) spd[34] * ftb - 500) / 1000; + /* Minimum CAS Latency Time (tAAmin) correction */ + dimm->tAA += (s32)((s8) spd[35] * ftb - 500) / 1000; + /* Minimum RAS# to CAS# Delay Time (tRCDmin) correction */ + dimm->tRCD += (s32)((s8) spd[36] * ftb - 500) / 1000; + /* Minimum Row Precharge Delay Time (tRPmin) correction */ + dimm->tRP += (s32)((s8) spd[37] * ftb - 500) / 1000; + /* Minimum Active to Active/Refresh Delay Time (tRCmin) corr. */ + dimm->tRC += (s32)((s8) spd[38] * ftb - 500) / 1000; + } + else { + printram(" no\n"); + } + /* SDRAM Optional Features */ reg8 = spd[30]; printram(" Optional features :"); |